The credibility estimates of the risk of several groups (schemes, fleets, regions, ...) from their experience over several periods, each weighted by its volume: each group's own mean, shrunk towards the collective by its credibility.
Value
A list with by_group (each group's weight, number of periods, mean,
credibility and estimate), the collective mean, the within and between variances
(s2, a) and k \(= s^2 / a\).
Details
With \(X_{ij}\) the value (such as a loss ratio or a claim frequency) of group \(i\) in period \(j\) and \(w_{ij}\) its weight (premium, exposure), \(w_i = \sum_j w_{ij}\), \(\bar X_i\) the weighted mean of the group and \(\bar X\) of all, the within-group variance is estimated by $$s^2 = \frac{\sum_i \sum_j w_{ij} (X_{ij} - \bar X_i)^2}{\sum_i (n_i - 1)}$$ and the variance between the groups by $$a = \frac{\sum_i w_i (\bar X_i - \bar X)^2 - (I - 1) s^2}{w - \sum_i w_i^2 / w},$$ with \(w = \sum_i w_i\). A group's credibility is \(Z_i = w_i / (w_i + s^2 / a)\) (0 for all when \(a \le 0\): the groups do not differ more than chance), the collective mean \(\hat\mu = \sum_i Z_i \bar X_i / \sum_i Z_i\) (the weighted mean when every Z is 0), and its estimate \(\hat\mu + Z_i (\bar X_i - \hat\mu)\).
References
Buhlmann, H. and Straub, E. (1970). Glaubwurdigkeit fur Schadensatze. Mitteilungen der Vereinigung Schweizerischer Versicherungsmathematiker 70, 111-133.
See also
credibility_weight, limited fluctuation credibility.
Examples
set.seed(1)
schemes <- data.frame(scheme = rep(c("A", "B", "C", "D"), each = 5),
premium = round(runif(20, 1e5, 1e6)))
true <- c(A = 0.55, B = 0.65, C = 0.70, D = 0.80)[schemes$scheme]
schemes$loss_ratio <- rgamma(20, shape = schemes$premium / 2e4, rate = schemes$premium / 2e4 / true)
buhlmann_straub(schemes$loss_ratio, schemes$premium, schemes$scheme)$by_group
#> group weight periods mean credibility estimate
#> 1 A 2588339 5 0.5523689 0.5428446 0.6017959
#> 2 B 3375288 5 0.6839754 0.6076069 0.6747589
#> 3 C 2501149 5 0.6774360 0.5343293 0.6695435
#> 4 D 3528234 5 0.7176997 0.6181211 0.6958515
